EP3776941A1 - Learning in communication systems - Google Patents
Learning in communication systemsInfo
- Publication number
- EP3776941A1 EP3776941A1 EP18774108.7A EP18774108A EP3776941A1 EP 3776941 A1 EP3776941 A1 EP 3776941A1 EP 18774108 A EP18774108 A EP 18774108A EP 3776941 A1 EP3776941 A1 EP 3776941A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- transmitter
- receiver
- training sequence
- algorithm
- loss function
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/02—Transmitters
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/06—Receivers
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L1/00—Arrangements for detecting or preventing errors in the information received
- H04L1/004—Arrangements for detecting or preventing errors in the information received by using forward error control
- H04L1/0041—Arrangements at the transmitter end
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L1/00—Arrangements for detecting or preventing errors in the information received
- H04L1/004—Arrangements for detecting or preventing errors in the information received by using forward error control
- H04L1/0045—Arrangements at the receiver end
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/0202—Channel estimation
- H04L25/0204—Channel estimation of multiple channels
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/0202—Channel estimation
- H04L25/0224—Channel estimation using sounding signals
- H04L25/0226—Channel estimation using sounding signals sounding signals per se
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03006—Arrangements for removing intersymbol interference
- H04L25/03012—Arrangements for removing intersymbol interference operating in the time domain
- H04L25/03114—Arrangements for removing intersymbol interference operating in the time domain non-adaptive, i.e. not adjustable, manually adjustable, or adjustable only during the reception of special signals
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03006—Arrangements for removing intersymbol interference
- H04L25/03165—Arrangements for removing intersymbol interference using neural networks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03006—Arrangements for removing intersymbol interference
- H04L25/03343—Arrangements at the transmitter end
Definitions
- the present specification relates to learning in communication systems.
- a simple communications system includes a transmitter, a transmission channel, and a receiver.
- the design of such communications systems may involve the separate design and optimisation of each part of the system.
- An alternative approach is to consider the entire communication system as a single system and to seek to optimise the entire system.
- this specification describes an apparatus comprising: means for receiving a first receiver-training sequence of messages at a first receiver of a first module of a transmission system from a second transmitter over a channel, wherein the transmission system comprises the first module having a first transmitter and the first receiver, a second module having the second transmitter and a second receiver, and the channel, wherein the first transmitter includes a first transmitter algorithm having at least some trainable weights, the second transmitter includes a second transmitter algorithm having at least some trainable weights, the first receiver has a first receiver algorithm having at least some trainable weights, and the second receiver includes a second receiver algorithm having at least some trainable weights; means for obtaining or generating a first receiver loss function, the first receiver loss function being generated based on the first receiver-training sequence as received at the first receiver and knowledge of the first receiver-training sequence as transmitted by the second transmitter; means for training at least some weights of the first receiver training algorithm based on the first receiver loss function; means for transmitting a second receiver-training sequence of messages from the first transmitter
- the first and/or the second receiver algorithm may comprise a receiver neural network.
- the first and/or the second transmitter algorithm may comprise a transmitter neural network.
- Some embodiments include means for generating the perturbed version of the first transmitter-training sequence of messages at the first module of the transmission system.
- the perturbations may be zero-mean Gaussian perturbations.
- the first transmitter loss function may be generated at the second module.
- the first transmitter loss function may be based on a mean-squared error between the first transmitter-training sequence of messages and the perturbed version of the first 20 transmitter-training sequence of messages.
- Some embodiments include means for repeating the training of the at least some weights of the first transmitter algorithm of the first transmitter until a first condition is reached.
- the first condition may, for example, be a defined number of iterations 25 and/or a defined performance level.
- the means for training at least some weights of the first transmitter may further comprise optimising one or more of a batch size of the first transmitter-training sequence of messages, a learning rate, and a distribution of the 30 perturbations applied to the perturbed versions of the first transmitter-training
- Some embodiments include means for repeating the training of the at least some weights of the first receiver algorithm until a second condition is reached.
- the second 35 condition may be a defined number of iterations and/ or a defined performance level.
- the first receiver loss function may be based on a mean-squared error between the first receiver-training sequence of messages as transmitted by the second transmitter and the first receiver-training sequence of messages as received at the first receiver.
- Some embodiments include means for repeating both the training of the at least some weights of the first transmitter algorithm and repeating the training of the at least some weights of the first receiver algorithm until a third condition is reached.
- Some embodiments include: means for receiving a perturbed version of a second transmitter-training sequence of messages for the second transmitter at the first receiver; and means for generating a second loss function, wherein the second loss function is based on the received perturbed version of the second transmitter-training sequence of messages as received at the first receiver and knowledge of the second transmitter-training sequence of messages for the second transmitter of the transmission system. Furthermore, some embodiments include means for sending the second loss function from the first transmitter to the second receiver.
- Some embodiments include: means for receiving the perturbed version of the first transmitter-training sequence of messages for the first transmitter at the second receiver; means for generating the first transmitter loss function; and means for sending the first transmitter loss function from the second transmitter to the first receiver.
- the said means may comprise: at least one processor; and at least one memory including computer program code, the at least one memory and computer program code configured to, with the at least one processor, cause the performance of the apparatus.
- this specification describes a method comprising: receiving a first receiver-training sequence of messages at a first receiver of a first module of a transmission system from a second transmitter over a channel, wherein the transmission system comprises the first module having a first transmitter and the first receiver, a second module having the second transmitter and a second receiver, and the channel, wherein the first transmitter includes a first transmitter algorithm having at least some trainable weights, the second transmitter includes a second transmitter algorithm having at least some trainable weights, the first receiver has a first receiver algorithm having at least some trainable weights, and the second receiver includes a second receiver algorithm having at least some trainable weights; obtaining or generating a first receiver loss function, the first receiver loss function being generated based on the first receiver-training sequence as received at the first receiver and 5 knowledge of the first receiver-training sequence as transmitted by the second
- the first and/or the second receiver algorithm may comprise a receiver neural network.
- the first and/or the second transmitter algorithm may comprise a transmitter neural network.
- the perturbed version of the first transmitter-training sequence of messages may be 25 generated at the first module of the transmission system.
- the first transmitter loss function may be generated at the second module.
- the first transmitter loss function may be based on a mean-squared error between the 0 first transmitter-training sequence of messages and the perturbed version of the first transmitter-training sequence of messages.
- Some embodiments include repeating the training of the at least some weights of the first transmitter algorithm of the first transmitter until a first condition is reached.
- the 35 first condition may, for example, be a defined number of iterations and/or a defined performance level.
- Some embodiments include repeating the training of the at least some weights of the first receiver algorithm until a second condition is reached.
- the second condition maybe a defined number of iterations and/or a defined performance level.
- Some embodiments include means for repeating both the training of the at least some weights of the first transmitter algorithm and repeating the training of the at least 5 some weights of the first receiver algorithm until a third condition is reached.
- Some embodiments include: receiving a perturbed version of a second transmitter- training sequence of messages for the second transmitter at the first receiver; and generating a second loss function, wherein the second loss function is based on the w received perturbed version of the second transmitter-training sequence of messages as received at the first receiver and knowledge of the second transmitter-training sequence of messages for the second transmitter of the transmission system. Furthermore, some embodiments include sending the second loss function from the first transmitter to the second receiver.
- Some embodiments include: receiving the perturbed version of the first transmitter- training sequence of messages for the first transmitter at the second receiver;
- this specification describes an apparatus configured to perform any method as described with reference to the second aspect.
- this specification describes computer-readable instructions which,
- this specification describes a computer program comprising instructions for causing an apparatus to perform at least the following: receive a first 30 receiver-training sequence of messages at a first receiver of a first module of a
- the transmission system comprises the first module having a first transmitter and the first receiver, a second module having the second transmitter and a second receiver, and the channel
- the first transmitter includes a first transmitter algorithm having at 35 least some trainable weights
- the second transmitter includes a second transmitter algorithm having at least some trainable weights
- the first receiver has a first receiver algorithm having at least some trainable weights
- the second receiver includes a second receiver algorithm having at least some trainable weights
- obtain or generate a first receiver loss function the first receiver loss function being generated based on the first receiver-training sequence as received at the first receiver and knowledge of the first receiver-training sequence as transmitted by the second transmitter; train at least some weights of the first receiver training algorithm based on the first receiver loss function
- this specification describes a non-transitory computer-readable medium comprising program instructions stored thereon for performing at least the following: receiving a first receiver-training sequence of messages at a first receiver of a first module of a transmission system from a second transmitter over a channel, wherein the transmission system comprises the first module having a first transmitter and the first receiver, a second module having the second transmitter and a second receiver, and the channel, wherein the first transmitter includes a first transmitter algorithm having at least some trainable weights, the second transmitter includes a second transmitter algorithm having at least some trainable weights, the first receiver has a first receiver algorithm having at least some trainable weights, and the second receiver includes a second receiver algorithm having at least some trainable weights; obtaining or generating a first receiver loss function, the first receiver loss function being generated based on the first receiver-training sequence as received at the first receiver and knowledge of the first receiver-training sequence as transmitted by the second transmitter; training at least some weights of the first receiver training algorithm based on the first receiver loss function; transmitting
- FIG. 1 is a block diagram of an example end-to-end communication system in accordance with an example embodiment
- FIG. 2 is a flow chart showing an algorithm in accordance with an example
- FIG. 3 is a block diagram of a transmitter used in example embodiments
- FIG. 4 is a block diagram of a receiver used in example embodiments.
- FIG. 5 is a block diagram of an example communication system in accordance with an example embodiment
- FIG. 6 is a block diagram of an example communication system in accordance with an example embodiment
- FIG. 7 is a flow chart showing an algorithm in accordance with an example
- FIG. 8 is a flow chart showing an algorithm in accordance with an example
- FIG. 9 is a block diagram of components of a system in accordance with an exemplary embodiment.
- FIGS. 10a and 10b show tangible media, respectively a removable non-volatile memory unit and a compact disc (CD) storing computer-readable code which when run by a computer perform operations according to embodiments.
- CD compact disc
- FIG. l is a block diagram of an example end-to-end communication system, indicated generally by the reference numeral 10, in accordance with an example embodiment.
- the system l includes a first agent 12, a channel 14 and a second agent 16.
- the first and second agents 12 and 16 are sometimes referred to herein as first and second modules.
- the first agent or module 12 includes a first transmitter 18 and a first receiver 20.
- the second agent or module 16 includes a second receiver 22 and a second transmitter 24.
- the system 10 converts an first input symbol (s) (also called a message) at the input of the first transmitter 18 into a first output symbol (s) at the output of the second receiver 22 and converts a second input symbol s' at the input to the second transmitter 24 into a second output symbol at the output of the first receiver 20.
- the first transmitter 18 and the second transmitter 24 include modules (such as neural networks) for implementing transmitter algorithms.
- the first receiver 20 and the second receiver 22 include modules (such as neural networks) for implementing receiver algorithms. At least some of the modules may be trained in order to optimise the performance of the system 10. More specifically, one or more of the transmitter algorithms and/or one or more of the receiver algorithms may include at least some trainable weights (which may be trainable through stochastic gradient descent and/or reinforcement learning).
- This representation can be seen as complex-baseband symbols.
- the transmitter hardware imposes constraints on x, e.g. an energy constraint
- the channel is described by the conditional probability density function (pdf) p(y
- the receiver Upon reception of y . the receiver produces the estimate s of the transmitted real-valued vector s.
- the second transmitter 24 communicates a real-valued vector
- the second transmitter 24 sends a complex- valued representation x' - x(s') e C w of s through the channel 14.
- the transmitters 18 and 24 define the mapping: TX: R M ⁇ ® C N .
- the transmitters map a real-valued vector (s 6 R M in the case of the transmitter 18 and s' e R M in the case of the transmitter 24) to an N-dimensional complex-valued vector.
- This mapping can be implemented as a neural network, as described below. Other parametric differential mappings could be provided instead of a neural network.
- the receivers 20 and 22 define the mapping: RX: € N ® H3 ⁇ 4 M .
- the receivers map a complex-valued vector (y' e C w in the case of the receiver 20 and y e C N in the case of the receiver 22) to an N-dimensional real-valued vector.
- This mapping can be implemented as a neural network, as described below. Other parametric differential mappings could be provided instead of a neural network.
- FIG. 2 is a flow chart showing an algorithm, indicated generally by the reference numeral 30, in accordance with an example embodiment.
- the algorithm 30 starts at operation 32 where the first receiver 20 of the first module 12 is trained by making the second transmitter 24 send predefined data to the first receiver 20 over the channel 14 and using supervised learning techniques to optimise the first receiver 20.
- the second receiver 22 of the second module 16 is trained by making the first transmitter 18 send predefined data to the second receiver 22 over the channel 14 and using supervised learning techniques to optimise the second receiver 22.
- the operations 32 and 34 represent an iteration of receiver training for both the first module 12 and the second module 16.
- the first transmitter 18 of the first module 12 is trained by enabling that transmitter to explore the space of possible real-valued vectors to channel symbol mappings, by introducing small perturbations (e.g. through the addition of a Gaussian zero-mean noise).
- the second receiver 22 provides feedback to the first transmitter in the form of a scalar loss relevant to a certain performance metric. (The feedback is sent by the second transmitter 24 over the channel 14 and received by the first receiver 20.) Using reinforcement learning, the gradient of the loss can be estimated.
- the second transmitter 24 of the second module 16 is trained by enabling that transmitter to explore the space of possible real valued vectors to channel symbol mappings, by introducing small perturbations, with the first receiver 20 providing feedback to the first transmitter (via channel 14) in the form of a scalar loss relevant to a certain performance metric.
- the feedback is sent by the first transmitter 18 over the channel 14 and received by the second receiver 22.
- the gradient of the loss can be estimated.
- the operations 36 and 38 represent an iteration of transmitter training for both the first module 12 and the second module 16.
- operation 40 it is determined whether the algorithm 30 is complete. If so, the algorithm terminates. If not, the algorithm returns to operation 32.
- the operation 40 of the algorithm 30 may take many forms. These include stopping after a fixed number of training iterations, stopping when the error measurement has not decreased during a fixed number of iteration or stopping when the mean square error has reached a desired value.
- the algorithm 30 can be used to implement multiple iterations of receiver and transmitter training for the system 10. Following the training process, the respective transmitters and receivers can communicate.
- FIG. 3 is a block diagram of a transmitter, indicated generally by the reference numeral 50, used in example embodiments (such as the transmitters 18 and 24 described above).
- the transmitter 50 includes a dense layer of one or more units 52 (e.g. one or more neural networks), a complex vector generator 54 and a normalization module 56.
- the modules within the transmitter 50 are provided by way of example and modifications are possible.
- the complex vector generator 54 and the normalization module 56 could be provided in a different order.
- the message index s is fed into a succession of layers, e.g. dense neural network (NN) layers 52 with possible different non-linear activation functions (such as ReLU, tanh, signmoid, linear etc.).
- a normalization is applied by the normalization module 56 that ensures that power, amplitude or other constraints are met.
- the result of the normalization process is the transmit vector x of the transmitter 50 (where x e C N ). As noted above, the order of the complex vector generation and the normalization could be reversed.
- FIG. 4 is a block diagram of a receiver, indicated generally by the reference numeral 60, used in example embodiments (such as the receivers 20 and 22).
- the receiver 60 includes a real vector generator 62 and one or more layers 64 (e.g. one or more neural networks).
- the modules within the receiver 60 are provided by way of example and modifications are possible.
- the result is fed into the one or more layers 64, which layers may have different activation functions such as ReLU, tanh, sigmoid, linear, etc.
- the last layer has M output dimensions. This generates a M-dimension real-valued vector s e K M , which can be interpreted as a guess of s. It is also possible to use complex-valued neutral networks to avoid the mapping R2C.
- the receiver 6o maps an LG-dimensional complex-valued vector to an M- real-valued vector.
- Training of a communication system in accordance with the principles described herein can be considered to consist of two separate phases, one for the receivers of the w transmission system and one for the transmitters of the transmission system.
- the operations 32 and 34 of the algorithm 30 can be considered to be a first phase and the operations 36 and 38 of the algorithm 30 can be considered to be a second phase.
- FIG. 5 is a block diagram of an example communication system, indicated generally by 15 the reference numeral 70, illustrating a receiver training algorithm in accordance with an example embodiment.
- the system 70 demonstrates aspects of the receiver-training operations described above with reference to the operations 32 and 34 (i.e. the‘first phase’ described above).
- the system 70 includes the first transmitter 18 and the first receiver 20 of the first module 12 described above, the second receiver 22 and the second transmitter 24 of the second module 16 described above, and the channel 14 described above.
- the system 70 additionally comprises a training algorithm, indicated generally by the reference 25 numeral 72, and a loss function generation module, indicated generally by the reference numeral 74.
- the two modules (or agents) that are aiming to learn to communicate are denoted by A and 0 B, a e ⁇ A, B ⁇ denotes one of the two agents, and a is the other agent, such that:
- the transmitter of a (the first transmitter 18 in the example system 70) send a sequence of known real-valued vectors to the receiver of a (the receiver 22 of the system 70).
- the neural network of the receiver 22 is optimised (using training algorithm 72) based on stochastic gradient descent and reinforcement learning. The goal of the optimisation is to reduce the mean square error, defined in module 74 as:
- iIoO N R is the number of real-value training examples used
- ⁇ T is the symbol received.
- FIG. 6 is a block diagram of an example communication system, indicated generally by
- the system 90 demonstrates aspects of the transmitter training operations described above with reference to the operations 36 and 38 (i.e. the ‘second phase’ described above). 0
- the system 90 includes the first transmitter 18 and the first receiver 20 of the first module 12 described above, the second receiver 22 and the second transmitter 24 of the second module 16 described above, and the channel 14 described above.
- the system 90 additionally comprises a perturbation module 92 between the second transmitter 24 and the channel 14, a loss function generation module, indicated generally by the
- the transmitter of a (the second transmitter 24 of the system 90) sends a sequence of known real-valued vectors to the receiver 20.
- the transmitter signals associated with each vector are slightly perturbed by perturbation module 92 by adding random vectors e taken from a known distribution (e.g. zero mean Gaussian perturbations).
- the loss function generation module 94 computes a loss function based on the squared error for each received vector. No optimisation of the receiver 20 is carried out at this step.
- the agent a uses the first transmitter 18 to feedback the squared errors generated in loss function generation module 94 to the agent a, which agent receives the data through the second receiver 22 (from the channel 14).
- the agent a optimises the neural network of the second transmitter 24 by estimating the gradient of the loss with respect to its trainable parameters using the knowledge of the transmitted vectors and signals, as well as the distribution of the perturbations applied by the perturbation module 92.
- FIG. 7 is a flow chart showing an algorithm, indicated generally by the reference numeral 110, in accordance with an example embodiment, that provides further details of the receiver-training algorithm described above.
- the output vectors are transmitted over the channel 14 by transmitter 18.
- the agent a computes the mean square error (using module 74) from ST and ST.
- the agent a using the training algorithm 72 to optimise the receiver 22 by applying one step of a stochastic gradient describes to the trainable parameters of the receiver, using mean square error (MSE) as the loss function, where (as noted above):
- operation 122 it is determined whether the algorithm 110 is complete. If so, the algorithm terminates; otherwise, the algorithm returns to operation 112 and the operations 112 to 122 are repeated.
- the receiver of the agent a should know the set of transmitted messages ST non-causally. This can be achieved in many ways. For example, both agents may use the same random number generator initialised with the same seed. Alternatively, both agents could agree prior to training on the sets of S T which are to be used in each training iteration.
- the batch size N R as well as the learning rate (and possible other parameters of the chosen stochastic gradient descent variant e.g. ADAM, RMSProp, Momentum) could be optimization parameters of the training operation.
- the operation 122 of the algorithm 110 may take many forms. These include stopping after a fixed number of training iterations, stopping when the mean square error has not decreased during a fixed number of iteration or stopping when the mean square error has reached a desired value.
- FIG. 8 is a flow chart showing an algorithm, indicated generally by the reference numeral 130, in accordance with an example embodiment, that provides further details of the transmitter-training algorithm described above.
- the output vectors are output by the transmitter 24 to the perturbation module 92.
- r(e) could be the multivariate complex Gaussian distribution
- agent a transmits the perturbed vectors to the agent a over the channel 14 to the first receiver 20.
- module 94 of the agent a computes the mean square error LT, as follows:
- the mean square error is encoded as X L for transmission using the first transmitter 18.
- the encoded error function is transmitter from the transmitter 18 to the second receiver 22 in operation 144 and received and decoded at the receiver 22 in operation 146.
- the agent a uses the training algorithm 96 to optimise the second transmitter 24 by applying one step of a stochastic gradient describes to the trainable parameters of the transmitter, using mean square error (MSE) as the loss function, where:
- the loss function is a function of which the gradient with respect to the trainable parameters Q of the transmitter are computed.
- the function ⁇ eLr is also known as the policy gradient.
- operation 150 it is determined whether the algorithm 130 is complete. If so, the algorithm terminates; otherwise, the algorithm returns to operation 132 and the operations 132 to 150 are repeated.
- the receiver of the agent a should know the set of transmitted messages ST non-causally. This can be achieved in many ways. For example, both agents may use the same random number generator initialised with the same seed. Alternatively, both agents could agree prior to training on the sets of ST which are to be used in each training iteration.
- the batch size NT as well as the learning rate (and possible other parameters of the chosen stochastic gradient descent variant e.g. ADAM, RMSProp, Momentum) could be optimization parameters of the training operation.
- the operation 150 of the algorithm 130 may take many forms. These include stopping after a fixed number of training iterations, stopping when the mean square error has not decreased during a fixed number of iteration or stopping when the mean square error has reached a desired value.
- the training processes described herein encompass a number of variants.
- the use of reinforcement learning as described herein relies on exploring the policy space (i.e. the space of possible state to action mappings).
- the policy is the mapping implemented by the transmitter
- the state space is the source symbol alphabet
- the action space is C n .
- Exploring can be done in numerous ways, two of the most popular approaches being: ⁇ Gaussian policy, in which a perturbation vector e is drawn from a multivariate zero- mean normal distribution and added to the current policy. This ensures exploration “in the neighbourhood” of the current policy.
- FIG. 9 is a schematic diagram of components of one or more of the example embodiments described previously, which hereafter are referred to generically as processing systems 300.
- a processing system 300 may have a processor 302, a memory 304 closely coupled to the processor and comprised of a RAM 314 and ROM 312, and, optionally, user input 310 and a display 318.
- the processing system 300 may comprise one or more network/apparatus interfaces 308 for connection to a network/apparatus, e.g. a modem which may be wired or wireless. Interface 308 may also operate as a connection to other apparatus such as device/apparatus which is not network side apparatus. Thus direct connection between devices/apparatus without network participation is possible.
- the processor 302 is connected to each of the other components in order to control operation thereof.
- the memory 304 may comprise a non-volatile memory, such as a hard disk drive (HDD) or a solid state drive (SSD).
- the ROM 312 of the memory 314 stores, amongst other things, an operating system 315 and may store software applications 316.
- the RAM 314 of the memory 304 is used by the processor 302 for the temporary storage of data.
- the operating system 315 may contain code which, when executed by the processor implements aspects of the algorithms 30, 110 or 130 described above. Note that in the case of small device/apparatus the memory can be most suitable for small size usage i.e. not always hard disk drive (HDD) or solid state drive (SSD) is used.
- the processor 302 may take any suitable form. For instance, it may be a
- microcontroller a plurality of microcontrollers, a processor, or a plurality of processors.
- the processing system 300 may be a standalone computer, a server, a console, or a network thereof.
- the processing system 300 and needed structural parts may be all inside device/apparatus such as IoT device/apparatus i.e. embedded to very small size
- the processing system 300 may also be associated with external software applications. These maybe applications stored on a remote server device/apparatus and may run partly or exclusively on the remote server
- the processing system 300 may be in communication with the remote server
- Figures 10a and 10b show tangible media, respectively a removable non-volatile memory unit 365 and a compact disc (CD) 368, storing computer-readable code which when run by a computer may perform methods according to example embodiments described above.
- the removable memory unit 365 may be a memory stick, e.g. a USB memory stick, having internal memory 366 storing the computer-readable code.
- the memory 366 may be accessed by a computer system via a connector 367.
- the CD 368 may be a CD-ROM or a DVD or similar. Other forms of tangible storage media may be used. Tangible media can be any device/apparatus capable of storing
- Embodiments of the present invention may be implemented in software, hardware, application logic or a combination of software, hardware and application logic.
- the software, application logic and/or hardware may reside on memory, or any computer media.
- the application logic, software or an instruction set is maintained on any one of various conventional computer-readable media.
- a“memory” or“computer-readable medium” may be any non-transitory media or means that can contain, store, communicate, propagate or transport the instructions for use by or in connection with an instruction execution system, apparatus, or device, such as a computer.
- references to, where relevant,“computer-readable storage medium”,“computer program product”,“tangibly embodied computer program” etc., or a“processor” or “processing circuitry” etc. should be understood to encompass not only computers having differing architectures such as single/multi-processor architectures and sequencers/parallel architectures, but also specialised circuits such as field programmable gate arrays FPGA, application specify circuits ASIC, signal processing devices/apparatus and other devices/apparatus. References to computer program, instructions, code etc.
- programmable processor firmware such as the programmable content of a hardware device/apparatus as instructions for a processor or configured or configuration settings for a fixed function device/apparatus, gate array, programmable logic device/apparatus, etc.
- circuitry refers to all of the following: (a) hardware-only circuit implementations (such as implementations in only analogue and/or digital circuitry) and (b) to combinations of circuits and software (and/or firmware), such as (as applicable): (i) to a combination of processor(s) or (ii) to portions of processor(s) /software (including digital signal processor(s)), software, and memory(ies) that work together to cause an apparatus, such as a server, to perform various functions) and (c) to circuits, such as a microprocessor(s) or a portion of a microprocessor(s), that require software or firmware for operation, even if the software or firmware is not physically present.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Power Engineering (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Mobile Radio Communication Systems (AREA)
- Information Transfer Between Computers (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/IB2018/000814 WO2019193377A1 (en) | 2018-04-03 | 2018-04-03 | End-to-end learning in communication systems |
PCT/IB2018/001029 WO2019193380A1 (en) | 2018-04-03 | 2018-06-29 | Learning in communication systems |
Publications (1)
Publication Number | Publication Date |
---|---|
EP3776941A1 true EP3776941A1 (en) | 2021-02-17 |
Family
ID=63517920
Family Applications (3)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP18765704.4A Pending EP3776365A1 (en) | 2018-04-03 | 2018-04-03 | End-to-end learning in communication systems |
EP18730331.8A Pending EP3776940A1 (en) | 2018-04-03 | 2018-06-07 | Learning in communication systems |
EP18774108.7A Pending EP3776941A1 (en) | 2018-04-03 | 2018-06-29 | Learning in communication systems |
Family Applications Before (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP18765704.4A Pending EP3776365A1 (en) | 2018-04-03 | 2018-04-03 | End-to-end learning in communication systems |
EP18730331.8A Pending EP3776940A1 (en) | 2018-04-03 | 2018-06-07 | Learning in communication systems |
Country Status (4)
Country | Link |
---|---|
US (3) | US20210201135A1 (en) |
EP (3) | EP3776365A1 (en) |
CN (3) | CN112236782A (en) |
WO (3) | WO2019193377A1 (en) |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112740631A (en) | 2018-07-20 | 2021-04-30 | 诺基亚技术有限公司 | Learning in a communication system by receiving updates of parameters in an algorithm |
CN113906704A (en) * | 2019-05-30 | 2022-01-07 | 诺基亚技术有限公司 | Learning in a communication system |
FI20195547A1 (en) | 2019-06-20 | 2020-12-21 | Nokia Technologies Oy | Communication system having a configurable modulation order and an associated method and apparatus |
US11570030B2 (en) * | 2019-10-11 | 2023-01-31 | University Of South Carolina | Method for non-linear distortion immune end-to-end learning with autoencoder—OFDM |
US11128498B2 (en) | 2020-02-25 | 2021-09-21 | Nokia Solutions And Networks Oy | Communication-channel tracking aided by reinforcement learning |
US20210303662A1 (en) * | 2020-03-31 | 2021-09-30 | Irdeto B.V. | Systems, methods, and storage media for creating secured transformed code from input code using a neural network to obscure a transformation function |
EP4154184A2 (en) * | 2020-05-22 | 2023-03-29 | Nokia Technologies Oy | Communication system |
EP3975453A1 (en) * | 2020-09-24 | 2022-03-30 | Rohde & Schwarz GmbH & Co. KG | Adaptive coverage optimization in single-frequency networks (sfn) |
CN113193925B (en) * | 2021-02-09 | 2023-08-11 | 中国人民解放军战略支援部队信息工程大学 | Optimization processing method and device of communication system and electronic equipment |
CN115186797B (en) * | 2022-06-06 | 2023-05-30 | 山西大学 | Method and system for communication of language |
CN117295096B (en) * | 2023-11-24 | 2024-02-09 | 武汉市豪迈电力自动化技术有限责任公司 | Smart electric meter data transmission method and system based on 5G short sharing |
Family Cites Families (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5293453A (en) * | 1990-06-07 | 1994-03-08 | Texas Instruments Incorporated | Error control codeword generating system and method based on a neural network |
JP4154229B2 (en) * | 2002-12-27 | 2008-09-24 | 富士通株式会社 | Adaptive array antenna controller |
GB2398976B (en) * | 2003-02-28 | 2006-03-08 | Samsung Electronics Co Ltd | Neural network decoder |
US7606138B2 (en) * | 2003-09-29 | 2009-10-20 | Her Majesty The Queen In Right Of Canada, As Represented By The Minister Of Industry, Through The Communications Research Centre Canada | Multi-symbol encapsulated OFDM system |
US7631246B2 (en) * | 2006-01-09 | 2009-12-08 | Broadcom Corporation | LDPC (low density parity check) code size adjustment by shortening and puncturing |
US8331420B2 (en) * | 2008-04-14 | 2012-12-11 | General Dynamics C4 Systems, Inc. | Methods and apparatus for multiple-antenna communication of wireless signals with embedded pilot signals |
JP5308102B2 (en) * | 2008-09-03 | 2013-10-09 | 日本電信電話株式会社 | Identification score / posterior probability calculation method by number of errors, error number weighted identification learning device using the method, method thereof, speech recognition device using the device, program, and recording medium |
EP2578051B1 (en) * | 2010-06-03 | 2018-10-24 | Nokia Solutions and Networks Oy | Base station calibration |
US9015093B1 (en) * | 2010-10-26 | 2015-04-21 | Michael Lamport Commons | Intelligent control with hierarchical stacked neural networks |
WO2014205376A1 (en) * | 2013-06-21 | 2014-12-24 | Marvell World Trade Ltd. | Methods and systems for determining indicators used in channel state information (csi) feedback in wireless systems |
EP3207726A1 (en) * | 2014-11-20 | 2017-08-23 | Huawei Technologies Co., Ltd. | Methods and nodes in a wireless communication network |
CN106295678B (en) * | 2016-07-27 | 2020-03-06 | 北京旷视科技有限公司 | Neural network training and constructing method and device and target detection method and device |
US10546230B2 (en) * | 2016-08-12 | 2020-01-28 | International Business Machines Corporation | Generating labeled data by sequence-to-sequence modeling with added perturbations to encoded information |
US10552738B2 (en) * | 2016-12-15 | 2020-02-04 | Google Llc | Adaptive channel coding using machine-learned models |
CN107241320A (en) * | 2017-05-26 | 2017-10-10 | 微梦创科网络科技(中国)有限公司 | A kind of man-machine discrimination method and identification system based on image |
-
2018
- 2018-04-03 WO PCT/IB2018/000814 patent/WO2019193377A1/en unknown
- 2018-04-03 CN CN201880094274.8A patent/CN112236782A/en active Pending
- 2018-04-03 US US17/044,607 patent/US20210201135A1/en active Pending
- 2018-04-03 EP EP18765704.4A patent/EP3776365A1/en active Pending
- 2018-06-07 US US17/044,584 patent/US20210027161A1/en active Pending
- 2018-06-07 WO PCT/EP2018/065006 patent/WO2019192740A1/en unknown
- 2018-06-07 EP EP18730331.8A patent/EP3776940A1/en active Pending
- 2018-06-07 CN CN201880093531.6A patent/CN112166568B/en active Active
- 2018-06-29 WO PCT/IB2018/001029 patent/WO2019193380A1/en unknown
- 2018-06-29 US US17/044,553 patent/US11082264B2/en active Active
- 2018-06-29 EP EP18774108.7A patent/EP3776941A1/en active Pending
- 2018-06-29 CN CN201880093390.8A patent/CN112166567B/en active Active
Also Published As
Publication number | Publication date |
---|---|
US20210201135A1 (en) | 2021-07-01 |
WO2019193380A1 (en) | 2019-10-10 |
WO2019193377A1 (en) | 2019-10-10 |
CN112166568A (en) | 2021-01-01 |
CN112166567B (en) | 2023-04-18 |
EP3776940A1 (en) | 2021-02-17 |
US20210099327A1 (en) | 2021-04-01 |
CN112236782A (en) | 2021-01-15 |
US20210027161A1 (en) | 2021-01-28 |
EP3776365A1 (en) | 2021-02-17 |
CN112166567A (en) | 2021-01-01 |
WO2019192740A1 (en) | 2019-10-10 |
US11082264B2 (en) | 2021-08-03 |
CN112166568B (en) | 2023-10-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP3776941A1 (en) | Learning in communication systems | |
US11556799B2 (en) | Channel modelling in a data transmission system | |
KR102620551B1 (en) | Learning in Communication Systems | |
US20210374529A1 (en) | End-to-end learning in communication systems | |
WO2019080988A1 (en) | End-to-end learning in communication systems | |
US20220209888A1 (en) | Iterative Detection in a Communication System | |
US11552731B2 (en) | Learning in communication systems by updating of parameters in a receiving algorithm | |
US20220083870A1 (en) | Training in Communication Systems | |
US20220263596A1 (en) | Transmitter Algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
17P | Request for examination filed |
Effective date: 20201103 |
|
AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
AX | Request for extension of the european patent |
Extension state: BA ME |
|
DAV | Request for validation of the european patent (deleted) | ||
DAX | Request for extension of the european patent (deleted) | ||
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: EXAMINATION IS IN PROGRESS |
|
17Q | First examination report despatched |
Effective date: 20221012 |